Back to Skills

ce-compound

Document a recently solved problem to compound your team's knowledge or CONCEPTS.md, the project's shared domain vocabulary.

21,938stars1,621forksUpdated 6/24/2026

Security Assessment

Medium Risk(60/100)

Detected risks:

Privilege Escalation([references/yaml-schema.md] sudo)
Security Score60/100

About ce-compound

A documentation command that captures a recently solved problem while context is fresh so a team's knowledge compounds over time. It coordinates multiple subagents working in parallel to produce structured documentation in docs/solutions/ with YAML frontmatter for searchability and future reference. The premise is that the first time a problem is solved takes research, but once documented the next occurrence takes minutes, so each documented solution compounds the team's knowledge.

It supports both interactive and headless invocation. The default interactive mode asks the user to choose Full versus Lightweight, asks whether to search session history (Full only), prompts for consent on a Discoverability Check, and ends with a "What's next?" menu, presenting blocking questions through the platform's question tool. Headless mode, intended for automations and skill-to-skill invocation where no human is present, asks no blocking questions, runs Full mode without session history, applies the Discoverability Check edit silently when a gap exists, skips specialized reviews, and ends with a structured terminal report; the resulting document is identical to what an interactive Full run would produce. Mode is detected by a mode:headless token in the arguments, which is stripped before the remainder is treated as a brief context hint. Full mode researches, cross-references, and reviews the solution; Lightweight produces the same documentation in a single pass without duplicate detection or cross-referencing, suited to simple fixes or long sessions near context limits.

The primary deliverable is a single file: the final solution documentation. Phase 1 subagents return text data to the orchestrator and must not write files; only the orchestrator writes. Beyond the solution doc it may make small maintenance side-effect writes, namely creating or updating CONCEPTS.md (the project's shared domain vocabulary) in a Vocabulary Capture step when a qualifying term surfaces, and a small edit to an AGENTS.md or CLAUDE.md instruction file when the Discoverability Check finds a gap. A standalone request to bootstrap CONCEPTS.md from scratch is redirected to ce-compound-refresh, since this command only seeds the learning's area as a side effect.

FAQ

What does this command produce?

Structured documentation of a recently solved problem written to docs/solutions/ with YAML frontmatter for searchability. The primary deliverable is a single file, the final solution doc, written only by the orchestrator.

What is the difference between Full and Lightweight mode?

Full mode researches, cross-references, and reviews the solution to detect duplicates and link existing docs. Lightweight produces the same documentation in a single pass without duplicate detection or cross-referencing, and is best for simple fixes or long sessions near context limits.

How does headless mode differ from interactive?

Headless mode asks no blocking questions, runs Full mode without session history, applies the Discoverability Check edit silently, skips specialized reviews, and ends with a structured terminal report. It is triggered by a mode:headless token in the arguments and is meant for automations.

Does it write files other than the solution doc?

Only maintenance side effects: it may create or update CONCEPTS.md when a qualifying domain term surfaces, and make a small edit to an AGENTS.md or CLAUDE.md instruction file when the Discoverability Check finds a gap. Phase 1 subagents return text and write nothing.

Can it bootstrap CONCEPTS.md from scratch?

No. It only seeds the learning's area as a side effect of documenting a real solution. A standalone request to build the repo-wide concept map is redirected to ce-compound-refresh.

All Files

17 files
references/agents/data-integrity-guardian.md3.1 KB
View
references/agents/performance-oracle.md4.5 KB
View
references/concepts-vocabulary.md6.1 KB
View
scripts/session-history/discover-sessions.sh4.2 KB
View
scripts/session-history/extract-skeleton.py20.2 KB
View
references/agents/best-practices-researcher.md7.3 KB
View
references/agents/pattern-recognition-specialist.md3.6 KB
View
references/agents/session-historian.md6.7 KB
View
references/yaml-schema.md5.7 KB
View
scripts/session-history/extract-metadata.py17.1 KB
View
SKILL.md50.1 KB
View
assets/resolution-template.md2.4 KB
View
references/agents/framework-docs-researcher.md5.5 KB
View
references/agents/security-sentinel.md4.0 KB
View
references/schema.yaml7.4 KB
View
scripts/session-history/extract-errors.py8.2 KB
View
scripts/validate-frontmatter.py4.9 KB
View

Install ce-compound

Download and extract the skill files to your .claude/skills/ directory.

Quick Setup:

  1. Copy the skill folder to .claude/skills/
  2. Claude will automatically detect and use the skill